FX Basics

Filtering

STOMPBOX DESIGN WORKSHOP

Esteban Maestre

CCRMA - Stanford University July 2012

FX Basics: Filtering

Filtering effects modify the frequency content of the audio signal, achieving boosting or weakening specific frequency bands or regions. Although their broad application to processing sound signals dates back from the early days of recording, their use application to processing guitar electrical signal may have started in the 1950s. Filtering effects make use of filters, which are signal processors which alter magnitude and phase of signals by different amount s to different frequency components. Ex: equalization, wah-wah

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

Equalization FX Basics:

Filtering

Original term coined from the task of ‘adjusting the balance between of (or equalize)’ different frequency components of a signal.

Equalization is commonly achieved by means of a device specifically designed for a user-friendly control of the parameters governing the behavior of filters used for its construction.

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

frequency

magnitude

CURRENT FREQUENCY CONTENT

DESIRED FREQUENCY CONTENT

User-friendly interface to controlling filters so that a desired alteration is achieved… DIGITAL FILTERS!

Digital Filters FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

DIGITAL FILTER Output

Signal

y[n] x[n]

Input Signal

Systems that perform mathematical operations (multiplications and additions) to a discrete input signal x[n] to modify some of its characteristics and obtain a discrete output signal y[n].

It is common to describe a digital filter in terms of how it affects amplitude and phase of different frequency components of a signal.

Ultimately, the design of digital filters is driven by such desired features. In general, digital filter design is not an easy task.

Digital Filters (ii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

Magnitude Response

Phase Response

fs/2 fc

1

90

LOWER FREQUENCIES ARE UNALTERED

Characteristic Frequency

HIGHER FREQUENCIES ARE ATTENUATED BY A FACTOR OF 2

0.5

HIGHER FREQUENCIES ARE DELAYED BY A QUARTER OF A PERIOD

fL fH

1/fL

1/fH

90 deg

Digital Filters (iii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

h[n] y[n] = x[n]*h[n] x[n]

f

|H(f)|

FREQUENCY RESPONSE

FREQUENY DOMAIN

IMPULSE RESPONSE

TIME DOMAIN

<H(f)

h[n]

n

DFT

H[f] Y[f] = X[f]·H[f] X[f]

fs/2

CONVOLUTION PRODUCT

δ[n] h[n]

IMPULSE (‘delta’ function)

IMPULSE RESPONSE

Digital Filters (iv) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

How to explore the frequency domain response of a given filter? Among other options…

SINUSOIDAL ANALYSIS

- Generate a sinusoidal xi[n] for each frequency fi to study

- Feed filter with each sinusoidal signal xi[n] and obtain a sinusoidal yi[n]

- Obtain magnitude and phase responses for each frequency fi :

|H(fi)| = A(yi)/A(xi) <H[fi] = <yi - <xi

IMPULSE RESPONSE

- Generate an impulse ‘delta’ signal δ[n]

- Feed filter with signal δ[n] and obtain output signal h[n]

- Obtain H(f) via DFT( h[n] )

- Obtain magnitude and phase responses as:

|H[f]| = abs( H[Ω] ) <H[Ω] = angle( H[Ω] )

Digital Filters (v) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

bi , aj FILTER COEFFICIENTS

Digital filters are commonly expressed by their difference equation:

y[n] = b0·x[n] + b1·x[n-1] + … + bM·x[n-M]

- a1·y[n-1] - … - aN·y[n-N]

i=0

M N

j=1 = Σ bi·x[n-i] - Σ aj·y[n-j]

CURRENT AND PREVIOUS INPUT SAMPLES

PREVIOUS OUTPUT SAMPLES

NON-RECURSIVE PART

RECURSIVE PART

max(M,N) FILTER ORDER

…or by their transfer function (in the frequency domain, through the ‘Z’ transform):

H(z) = = b0 + b1·z-1 +… + bN·z-M

1+ a1·z-1 +… + aN·z-N

Y(z)

X(z) z-M denotes M samples of delay

Digital Filters (vi) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

Two main types of digital filters:

Finite Impulse Response ( FIR )

- Presents only bi coefficients being non-zero : NON-RECURSIVE

- Finite h[n]

- Phase response is linear

H(z) =

b0 + b1·z-1 +… + bN·z-M

1+ a1·z-1 +… + aN·z-N

Infinite Impulse Response ( IIR )

- Presents both bi and aj coefficients being non-zero: RECURSIVE

- Infinite h[n]

- Phase response is non-linear

- Need less computations for similar desired characteristics

- May suffer from numerical problems due to feedback

Digital Filters (vii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

Some prototypical basic filters ( magnitude response ):

ALL-PASS (AP)

Freq. (log)

Gai

n (

log)

LOW-PASS (LP)

f0

HIGH-PASS (HP)

f0

BAND-PASS (BP)

fL fH

HIGH-SHELF

f0

GH

LOW-SHELF

f0

GL

NOTCH

f0

BW

PEAK

f0

BW

Introduces a desired PHASE SHIFT

Digital Filters (viii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

LPF (Butterworth) design parameters/constraints:

f (oct)

dB

f0

f (oct)

dB

f0

1st ORDER

2nd ORDER

-6dB/oct Roll-Off

-12dB/oct Roll-Off

Characteristic Frequency

Low BW ( Bandwidth )

High Q ( Quality Factor )

Low Q

High BW

Desired {f0}

DESIGN PROCEDURE

Desired {f0,Q} or {f0,BW}

DESIGN PROCEDURE

Filter Coefficients {bi,aj}

Filter Coefficients {bi,aj}

Digital Filters (ix) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

BPF design parameters/constraints:

f (oct)

dB

f0

-12dB/oct Roll-Off

Desired {fL,fH} or {f0,BW}

Filter Coefficients {bi,aj}

2nd ORDER

fH fL

BW

PEAK design parameters/constraints:

f (oct)

dB

f0

Desired {f0,G0,BW} or {f0,G0,Q}

Filter Coefficients {bi,aj}

2nd ORDER

G0

DESIGN PROCEDURE

Q

BW

DESIGN PROCEDURE

Digital Filters (x) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

HIGH-SHELF design parameters/constraints:

f (oct)

dB

f0

Desired {f0,G0,S}

Filter Coefficients {bi,aj}

2nd ORDER

DESIGN PROCEDURE

G0

SHELF SLOPE +12dB/oct (max)

S

http://www.musicdsp.org/files/Audio-EQ-Cookbook.txt

04_stomp_filtering_1.pd

H(z) =

b0 + b1·z-1 + b2·z-2

a0 + a1·z-1 + aN·z-2

All these filters functions can be implemented by means of the 2nd order ‘BIQUAD’ section:

How to design them? Extensive theory & literature!! Quick method: R. Bristow-Johnson’s cookbook:

Equalization (ii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

N-BAND EQUALIZER by PARALLEL BAND-DEDICATED, FIXED FILTERS

LP BP BP BP HP … BP …

G G G G G G Variable Gains

Band Filters

… 0dB

+ GRAPHIC EQUALIZER Fixed Filters: only Control of Combination (Gains)

Input Signal

Output Signal

Equalization (iii) FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

N-BAND EQUALIZER by CASCADE of BAND-DEDICATED, CONTROLLABLE SECTIONS

Section 1 Section N … Input Signal Output Signal

0dB

+GMAX

LOW SHELF

fLOW fPEAK fHIGH

PEAK HIGH SHELF

GLOW|fLOW

PARAMETRIC EQUALIZER Variable Gains, Frequencies, and Bandwidths.

ISSUE: Changes in Gain/Frequency lead to Q/BW variation…

Constant-Q filters!

http://www.rane.com/note101.html

GMID|fMID|QMID

GHIGH|fHIGH

Wah-wah FX Basics:

Filtering

Stompbox Design Workshop July 2012 - CCRMA, Stanford University

Dating back from the 60s, its name was given after voice tone modulation (formant shift) caused by transition between vowels.

G

RESONANT FILTER

f0

Pedal Angle

http://www.geofex.com/article_folders/wahpedl/voicewah.htm

Input Signal

map In its most basic form, it consists on shifting the center frequency of a resonant filter (Peak BP or LP)

Output Signal

0dB

G

fMIN f0 fMAX

05_stomp_filtering_2.pd

fMIN f0 fMAX

0dB